]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
rxrpc: Move call state changes from recvmsg to I/O thread
authorDavid Howells <dhowells@redhat.com>
Wed, 26 Oct 2022 22:43:00 +0000 (23:43 +0100)
committerDavid Howells <dhowells@redhat.com>
Fri, 6 Jan 2023 09:43:33 +0000 (09:43 +0000)
Move the call state changes that are made in rxrpc_recvmsg() to the I/O
thread.  This means that, thenceforth, only the I/O thread does this and
the call state lock can be removed.

This requires the Rx phase to be ended when the last packet is received,
not when it is processed.

Since this now changes the rxrpc call state to SUCCEEDED before we've
consumed all the data from it, rxrpc_kernel_check_life() mustn't say the
call is dead until the recvmsg queue is empty (unless the call has failed).

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

fs/afs/rxrpc.c
net/rxrpc/af_rxrpc.c
net/rxrpc/ar-internal.h
net/rxrpc/input.c
net/rxrpc/recvmsg.c

index bd3830bc670064ef123da0e1330d65751dafdbb0..7817e2b860e5eb30a2c758d78a737615993bdd85 100644 (file)
@@ -909,6 +909,7 @@ int afs_extract_data(struct afs_call *call, bool want_more)
        ret = rxrpc_kernel_recv_data(net->socket, call->rxcall, iter,
                                     &call->iov_len, want_more, &remote_abort,
                                     &call->service_id);
+       trace_afs_receive_data(call, call->iter, want_more, ret);
        if (ret == 0 || ret == -EAGAIN)
                return ret;
 
index 61c30d0f67356d9fbc9e77ea155df75df1d3285a..cf200e4e0eae1879a5d11fff0341c590757acd1b 100644 (file)
@@ -373,13 +373,17 @@ EXPORT_SYMBOL(rxrpc_kernel_end_call);
  * @sock: The socket the call is on
  * @call: The call to check
  *
- * Allow a kernel service to find out whether a call is still alive -
- * ie. whether it has completed.
+ * Allow a kernel service to find out whether a call is still alive - whether
+ * it has completed successfully and all received data has been consumed.
  */
 bool rxrpc_kernel_check_life(const struct socket *sock,
                             const struct rxrpc_call *call)
 {
-       return !rxrpc_call_is_complete(call);
+       if (!rxrpc_call_is_complete(call))
+               return true;
+       if (call->completion != RXRPC_CALL_SUCCEEDED)
+               return false;
+       return !skb_queue_empty(&call->recvmsg_queue);
 }
 EXPORT_SYMBOL(rxrpc_kernel_check_life);
 
index 9e992487649c417c14d8b56e95a45721ac0602c0..8612734397361722e54cfc62803040d04e03f6b0 100644 (file)
@@ -545,7 +545,8 @@ enum rxrpc_call_flag {
        RXRPC_CALL_KERNEL,              /* The call was made by the kernel */
        RXRPC_CALL_UPGRADE,             /* Service upgrade was requested for the call */
        RXRPC_CALL_EXCLUSIVE,           /* The call uses a once-only connection */
-       RXRPC_CALL_RX_IS_IDLE,          /* Reception is idle - send an ACK */
+       RXRPC_CALL_RX_IS_IDLE,          /* recvmsg() is idle - send an ACK */
+       RXRPC_CALL_RECVMSG_READ_ALL,    /* recvmsg() read all of the received data */
 };
 
 /*
index bd69ff2d90826e3a5d4857d02c6f3eb1e6e563c9..6eb21425f41fe611aa47d5c691934a59502d3444 100644 (file)
@@ -319,6 +319,41 @@ static bool rxrpc_receiving_reply(struct rxrpc_call *call)
        return true;
 }
 
+/*
+ * End the packet reception phase.
+ */
+static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
+{
+       rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
+
+       _enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
+
+       trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
+
+       if (rxrpc_call_state(call) == RXRPC_CALL_CLIENT_RECV_REPLY)
+               rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
+
+       write_lock(&call->state_lock);
+
+       switch (call->state) {
+       case RXRPC_CALL_CLIENT_RECV_REPLY:
+               __rxrpc_call_completed(call);
+               write_unlock(&call->state_lock);
+               break;
+
+       case RXRPC_CALL_SERVER_RECV_REQUEST:
+               call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
+               call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
+               write_unlock(&call->state_lock);
+               rxrpc_propose_delay_ACK(call, serial,
+                                       rxrpc_propose_ack_processing_op);
+               break;
+       default:
+               write_unlock(&call->state_lock);
+               break;
+       }
+}
+
 static void rxrpc_input_update_ack_window(struct rxrpc_call *call,
                                          rxrpc_seq_t window, rxrpc_seq_t wtop)
 {
@@ -337,8 +372,9 @@ static void rxrpc_input_queue_data(struct rxrpc_call *call, struct sk_buff *skb,
 
        __skb_queue_tail(&call->recvmsg_queue, skb);
        rxrpc_input_update_ack_window(call, window, wtop);
-
        trace_rxrpc_receive(call, last ? why + 1 : why, sp->hdr.serial, sp->hdr.seq);
+       if (last)
+               rxrpc_end_rx_phase(call, sp->hdr.serial);
 }
 
 /*
index 7bf36a8839eccdf6c4cbcf3051248949a2a4b1db..dd54ceee7bcc8e3fbf4d517b95f6e01b38a0d41e 100644 (file)
@@ -100,42 +100,6 @@ static int rxrpc_recvmsg_term(struct rxrpc_call *call, struct msghdr *msg)
        return ret;
 }
 
-/*
- * End the packet reception phase.
- */
-static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
-{
-       rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
-
-       _enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
-
-       trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
-
-       if (rxrpc_call_state(call) == RXRPC_CALL_CLIENT_RECV_REPLY)
-               rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
-
-       write_lock(&call->state_lock);
-
-       switch (call->state) {
-       case RXRPC_CALL_CLIENT_RECV_REPLY:
-               __rxrpc_call_completed(call);
-               write_unlock(&call->state_lock);
-               rxrpc_poke_call(call, rxrpc_call_poke_complete);
-               break;
-
-       case RXRPC_CALL_SERVER_RECV_REQUEST:
-               call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
-               call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
-               write_unlock(&call->state_lock);
-               rxrpc_propose_delay_ACK(call, serial,
-                                       rxrpc_propose_ack_processing_op);
-               break;
-       default:
-               write_unlock(&call->state_lock);
-               break;
-       }
-}
-
 /*
  * Discard a packet we've used up and advance the Rx window by one.
  */
@@ -166,10 +130,9 @@ static void rxrpc_rotate_rx_window(struct rxrpc_call *call)
 
        trace_rxrpc_receive(call, last ? rxrpc_receive_rotate_last : rxrpc_receive_rotate,
                            serial, call->rx_consumed);
-       if (last) {
-               rxrpc_end_rx_phase(call, serial);
-               return;
-       }
+
+       if (last)
+               set_bit(RXRPC_CALL_RECVMSG_READ_ALL, &call->flags);
 
        /* Check to see if there's an ACK that needs sending. */
        acked = atomic_add_return(call->rx_consumed - old_consumed,
@@ -194,7 +157,8 @@ static int rxrpc_verify_data(struct rxrpc_call *call, struct sk_buff *skb)
 /*
  * Deliver messages to a call.  This keeps processing packets until the buffer
  * is filled and we find either more DATA (returns 0) or the end of the DATA
- * (returns 1).  If more packets are required, it returns -EAGAIN.
+ * (returns 1).  If more packets are required, it returns -EAGAIN and if the
+ * call has failed it returns -EIO.
  */
 static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
                              struct msghdr *msg, struct iov_iter *iter,
@@ -210,7 +174,13 @@ static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
        rx_pkt_offset = call->rx_pkt_offset;
        rx_pkt_len = call->rx_pkt_len;
 
-       if (rxrpc_call_state(call) >= RXRPC_CALL_SERVER_ACK_REQUEST) {
+       if (rxrpc_call_has_failed(call)) {
+               seq = lower_32_bits(atomic64_read(&call->ackr_window)) - 1;
+               ret = -EIO;
+               goto done;
+       }
+
+       if (test_bit(RXRPC_CALL_RECVMSG_READ_ALL, &call->flags)) {
                seq = lower_32_bits(atomic64_read(&call->ackr_window)) - 1;
                ret = 1;
                goto done;
@@ -234,14 +204,15 @@ static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
 
                if (rx_pkt_offset == 0) {
                        ret2 = rxrpc_verify_data(call, skb);
-                       rx_pkt_offset = sp->offset;
-                       rx_pkt_len = sp->len;
                        trace_rxrpc_recvdata(call, rxrpc_recvmsg_next, seq,
-                                            rx_pkt_offset, rx_pkt_len, ret2);
+                                            sp->offset, sp->len, ret2);
                        if (ret2 < 0) {
+                               kdebug("verify = %d", ret2);
                                ret = ret2;
                                goto out;
                        }
+                       rx_pkt_offset = sp->offset;
+                       rx_pkt_len = sp->len;
                } else {
                        trace_rxrpc_recvdata(call, rxrpc_recvmsg_cont, seq,
                                             rx_pkt_offset, rx_pkt_len, 0);
@@ -416,36 +387,36 @@ try_again:
                msg->msg_namelen = len;
        }
 
-       switch (rxrpc_call_state(call)) {
-       case RXRPC_CALL_CLIENT_RECV_REPLY:
-       case RXRPC_CALL_SERVER_RECV_REQUEST:
-       case RXRPC_CALL_SERVER_ACK_REQUEST:
-               ret = rxrpc_recvmsg_data(sock, call, msg, &msg->msg_iter, len,
-                                        flags, &copied);
-               if (ret == -EAGAIN)
-                       ret = 0;
-
-               if (!skb_queue_empty(&call->recvmsg_queue))
-                       rxrpc_notify_socket(call);
-               break;
-       default:
+       ret = rxrpc_recvmsg_data(sock, call, msg, &msg->msg_iter, len,
+                                flags, &copied);
+       if (ret == -EAGAIN)
                ret = 0;
-               break;
-       }
-
+       if (ret == -EIO)
+               goto call_failed;
        if (ret < 0)
                goto error_unlock_call;
 
-       if (rxrpc_call_is_complete(call)) {
-               ret = rxrpc_recvmsg_term(call, msg);
-               if (ret < 0)
-                       goto error_unlock_call;
-               if (!(flags & MSG_PEEK))
-                       rxrpc_release_call(rx, call);
-               msg->msg_flags |= MSG_EOR;
-               ret = 1;
-       }
+       if (rxrpc_call_is_complete(call) &&
+           skb_queue_empty(&call->recvmsg_queue))
+               goto call_complete;
+       if (rxrpc_call_has_failed(call))
+               goto call_failed;
 
+       rxrpc_notify_socket(call);
+       goto not_yet_complete;
+
+call_failed:
+       rxrpc_purge_queue(&call->recvmsg_queue);
+call_complete:
+       ret = rxrpc_recvmsg_term(call, msg);
+       if (ret < 0)
+               goto error_unlock_call;
+       if (!(flags & MSG_PEEK))
+               rxrpc_release_call(rx, call);
+       msg->msg_flags |= MSG_EOR;
+       ret = 1;
+
+not_yet_complete:
        if (ret == 0)
                msg->msg_flags |= MSG_MORE;
        else
@@ -508,49 +479,34 @@ int rxrpc_kernel_recv_data(struct socket *sock, struct rxrpc_call *call,
        size_t offset = 0;
        int ret;
 
-       _enter("{%d,%s},%zu,%d",
-              call->debug_id, rxrpc_call_states[call->state],
-              *_len, want_more);
-
-       ASSERTCMP(call->state, !=, RXRPC_CALL_SERVER_SECURING);
+       _enter("{%d},%zu,%d", call->debug_id, *_len, want_more);
 
        mutex_lock(&call->user_mutex);
 
-       switch (rxrpc_call_state(call)) {
-       case RXRPC_CALL_CLIENT_RECV_REPLY:
-       case RXRPC_CALL_SERVER_RECV_REQUEST:
-       case RXRPC_CALL_SERVER_ACK_REQUEST:
-               ret = rxrpc_recvmsg_data(sock, call, NULL, iter,
-                                        *_len, 0, &offset);
-               *_len -= offset;
-               if (ret < 0)
-                       goto out;
-
-               /* We can only reach here with a partially full buffer if we
-                * have reached the end of the data.  We must otherwise have a
-                * full buffer or have been given -EAGAIN.
-                */
-               if (ret == 1) {
-                       if (iov_iter_count(iter) > 0)
-                               goto short_data;
-                       if (!want_more)
-                               goto read_phase_complete;
-                       ret = 0;
-                       goto out;
-               }
-
-               if (!want_more)
-                       goto excess_data;
+       ret = rxrpc_recvmsg_data(sock, call, NULL, iter, *_len, 0, &offset);
+       *_len -= offset;
+       if (ret == -EIO)
+               goto call_failed;
+       if (ret < 0)
                goto out;
 
-       case RXRPC_CALL_COMPLETE:
-               goto call_complete;
-
-       default:
-               ret = -EINPROGRESS;
+       /* We can only reach here with a partially full buffer if we have
+        * reached the end of the data.  We must otherwise have a full buffer
+        * or have been given -EAGAIN.
+        */
+       if (ret == 1) {
+               if (iov_iter_count(iter) > 0)
+                       goto short_data;
+               if (!want_more)
+                       goto read_phase_complete;
+               ret = 0;
                goto out;
        }
 
+       if (!want_more)
+               goto excess_data;
+       goto out;
+
 read_phase_complete:
        ret = 1;
 out:
@@ -572,7 +528,7 @@ excess_data:
                          0, -EMSGSIZE);
        ret = -EMSGSIZE;
        goto out;
-call_complete:
+call_failed:
        *_abort = call->abort_code;
        ret = call->error;
        if (call->completion == RXRPC_CALL_SUCCEEDED) {